]> asedeno.scripts.mit.edu Git - linux.git/blob - drivers/nvdimm/namespace_devs.c
fm10k: bump driver version to match out-of-tree release
[linux.git] / drivers / nvdimm / namespace_devs.c
1 /*
2  * Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or modify
5  * it under the terms of version 2 of the GNU General Public License as
6  * published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful, but
9  * WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  */
13 #include <linux/module.h>
14 #include <linux/device.h>
15 #include <linux/sort.h>
16 #include <linux/slab.h>
17 #include <linux/list.h>
18 #include <linux/nd.h>
19 #include "nd-core.h"
20 #include "pmem.h"
21 #include "nd.h"
22
23 static void namespace_io_release(struct device *dev)
24 {
25         struct nd_namespace_io *nsio = to_nd_namespace_io(dev);
26
27         kfree(nsio);
28 }
29
30 static void namespace_pmem_release(struct device *dev)
31 {
32         struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
33         struct nd_region *nd_region = to_nd_region(dev->parent);
34
35         if (nspm->id >= 0)
36                 ida_simple_remove(&nd_region->ns_ida, nspm->id);
37         kfree(nspm->alt_name);
38         kfree(nspm->uuid);
39         kfree(nspm);
40 }
41
42 static void namespace_blk_release(struct device *dev)
43 {
44         struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
45         struct nd_region *nd_region = to_nd_region(dev->parent);
46
47         if (nsblk->id >= 0)
48                 ida_simple_remove(&nd_region->ns_ida, nsblk->id);
49         kfree(nsblk->alt_name);
50         kfree(nsblk->uuid);
51         kfree(nsblk->res);
52         kfree(nsblk);
53 }
54
55 static const struct device_type namespace_io_device_type = {
56         .name = "nd_namespace_io",
57         .release = namespace_io_release,
58 };
59
60 static const struct device_type namespace_pmem_device_type = {
61         .name = "nd_namespace_pmem",
62         .release = namespace_pmem_release,
63 };
64
65 static const struct device_type namespace_blk_device_type = {
66         .name = "nd_namespace_blk",
67         .release = namespace_blk_release,
68 };
69
70 static bool is_namespace_pmem(const struct device *dev)
71 {
72         return dev ? dev->type == &namespace_pmem_device_type : false;
73 }
74
75 static bool is_namespace_blk(const struct device *dev)
76 {
77         return dev ? dev->type == &namespace_blk_device_type : false;
78 }
79
80 static bool is_namespace_io(const struct device *dev)
81 {
82         return dev ? dev->type == &namespace_io_device_type : false;
83 }
84
85 static int is_uuid_busy(struct device *dev, void *data)
86 {
87         u8 *uuid1 = data, *uuid2 = NULL;
88
89         if (is_namespace_pmem(dev)) {
90                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
91
92                 uuid2 = nspm->uuid;
93         } else if (is_namespace_blk(dev)) {
94                 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
95
96                 uuid2 = nsblk->uuid;
97         } else if (is_nd_btt(dev)) {
98                 struct nd_btt *nd_btt = to_nd_btt(dev);
99
100                 uuid2 = nd_btt->uuid;
101         } else if (is_nd_pfn(dev)) {
102                 struct nd_pfn *nd_pfn = to_nd_pfn(dev);
103
104                 uuid2 = nd_pfn->uuid;
105         }
106
107         if (uuid2 && memcmp(uuid1, uuid2, NSLABEL_UUID_LEN) == 0)
108                 return -EBUSY;
109
110         return 0;
111 }
112
113 static int is_namespace_uuid_busy(struct device *dev, void *data)
114 {
115         if (is_nd_region(dev))
116                 return device_for_each_child(dev, data, is_uuid_busy);
117         return 0;
118 }
119
120 /**
121  * nd_is_uuid_unique - verify that no other namespace has @uuid
122  * @dev: any device on a nvdimm_bus
123  * @uuid: uuid to check
124  */
125 bool nd_is_uuid_unique(struct device *dev, u8 *uuid)
126 {
127         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
128
129         if (!nvdimm_bus)
130                 return false;
131         WARN_ON_ONCE(!is_nvdimm_bus_locked(&nvdimm_bus->dev));
132         if (device_for_each_child(&nvdimm_bus->dev, uuid,
133                                 is_namespace_uuid_busy) != 0)
134                 return false;
135         return true;
136 }
137
138 bool pmem_should_map_pages(struct device *dev)
139 {
140         struct nd_region *nd_region = to_nd_region(dev->parent);
141         struct nd_namespace_io *nsio;
142
143         if (!IS_ENABLED(CONFIG_ZONE_DEVICE))
144                 return false;
145
146         if (!test_bit(ND_REGION_PAGEMAP, &nd_region->flags))
147                 return false;
148
149         if (is_nd_pfn(dev) || is_nd_btt(dev))
150                 return false;
151
152         nsio = to_nd_namespace_io(dev);
153         if (region_intersects(nsio->res.start, resource_size(&nsio->res),
154                                 IORESOURCE_SYSTEM_RAM,
155                                 IORES_DESC_NONE) == REGION_MIXED)
156                 return false;
157
158         return ARCH_MEMREMAP_PMEM == MEMREMAP_WB;
159 }
160 EXPORT_SYMBOL(pmem_should_map_pages);
161
162 unsigned int pmem_sector_size(struct nd_namespace_common *ndns)
163 {
164         if (is_namespace_pmem(&ndns->dev)) {
165                 struct nd_namespace_pmem *nspm;
166
167                 nspm = to_nd_namespace_pmem(&ndns->dev);
168                 if (nspm->lbasize == 0 || nspm->lbasize == 512)
169                         /* default */;
170                 else if (nspm->lbasize == 4096)
171                         return 4096;
172                 else
173                         dev_WARN(&ndns->dev, "unsupported sector size: %ld\n",
174                                         nspm->lbasize);
175         }
176
177         /*
178          * There is no namespace label (is_namespace_io()), or the label
179          * indicates the default sector size.
180          */
181         return 512;
182 }
183 EXPORT_SYMBOL(pmem_sector_size);
184
185 const char *nvdimm_namespace_disk_name(struct nd_namespace_common *ndns,
186                 char *name)
187 {
188         struct nd_region *nd_region = to_nd_region(ndns->dev.parent);
189         const char *suffix = NULL;
190
191         if (ndns->claim && is_nd_btt(ndns->claim))
192                 suffix = "s";
193
194         if (is_namespace_pmem(&ndns->dev) || is_namespace_io(&ndns->dev)) {
195                 int nsidx = 0;
196
197                 if (is_namespace_pmem(&ndns->dev)) {
198                         struct nd_namespace_pmem *nspm;
199
200                         nspm = to_nd_namespace_pmem(&ndns->dev);
201                         nsidx = nspm->id;
202                 }
203
204                 if (nsidx)
205                         sprintf(name, "pmem%d.%d%s", nd_region->id, nsidx,
206                                         suffix ? suffix : "");
207                 else
208                         sprintf(name, "pmem%d%s", nd_region->id,
209                                         suffix ? suffix : "");
210         } else if (is_namespace_blk(&ndns->dev)) {
211                 struct nd_namespace_blk *nsblk;
212
213                 nsblk = to_nd_namespace_blk(&ndns->dev);
214                 sprintf(name, "ndblk%d.%d%s", nd_region->id, nsblk->id,
215                                 suffix ? suffix : "");
216         } else {
217                 return NULL;
218         }
219
220         return name;
221 }
222 EXPORT_SYMBOL(nvdimm_namespace_disk_name);
223
224 const u8 *nd_dev_to_uuid(struct device *dev)
225 {
226         static const u8 null_uuid[16];
227
228         if (!dev)
229                 return null_uuid;
230
231         if (is_namespace_pmem(dev)) {
232                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
233
234                 return nspm->uuid;
235         } else if (is_namespace_blk(dev)) {
236                 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
237
238                 return nsblk->uuid;
239         } else
240                 return null_uuid;
241 }
242 EXPORT_SYMBOL(nd_dev_to_uuid);
243
244 static ssize_t nstype_show(struct device *dev,
245                 struct device_attribute *attr, char *buf)
246 {
247         struct nd_region *nd_region = to_nd_region(dev->parent);
248
249         return sprintf(buf, "%d\n", nd_region_to_nstype(nd_region));
250 }
251 static DEVICE_ATTR_RO(nstype);
252
253 static ssize_t __alt_name_store(struct device *dev, const char *buf,
254                 const size_t len)
255 {
256         char *input, *pos, *alt_name, **ns_altname;
257         ssize_t rc;
258
259         if (is_namespace_pmem(dev)) {
260                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
261
262                 ns_altname = &nspm->alt_name;
263         } else if (is_namespace_blk(dev)) {
264                 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
265
266                 ns_altname = &nsblk->alt_name;
267         } else
268                 return -ENXIO;
269
270         if (dev->driver || to_ndns(dev)->claim)
271                 return -EBUSY;
272
273         input = kmemdup(buf, len + 1, GFP_KERNEL);
274         if (!input)
275                 return -ENOMEM;
276
277         input[len] = '\0';
278         pos = strim(input);
279         if (strlen(pos) + 1 > NSLABEL_NAME_LEN) {
280                 rc = -EINVAL;
281                 goto out;
282         }
283
284         alt_name = kzalloc(NSLABEL_NAME_LEN, GFP_KERNEL);
285         if (!alt_name) {
286                 rc = -ENOMEM;
287                 goto out;
288         }
289         kfree(*ns_altname);
290         *ns_altname = alt_name;
291         sprintf(*ns_altname, "%s", pos);
292         rc = len;
293
294 out:
295         kfree(input);
296         return rc;
297 }
298
299 static resource_size_t nd_namespace_blk_size(struct nd_namespace_blk *nsblk)
300 {
301         struct nd_region *nd_region = to_nd_region(nsblk->common.dev.parent);
302         struct nd_mapping *nd_mapping = &nd_region->mapping[0];
303         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
304         struct nd_label_id label_id;
305         resource_size_t size = 0;
306         struct resource *res;
307
308         if (!nsblk->uuid)
309                 return 0;
310         nd_label_gen_id(&label_id, nsblk->uuid, NSLABEL_FLAG_LOCAL);
311         for_each_dpa_resource(ndd, res)
312                 if (strcmp(res->name, label_id.id) == 0)
313                         size += resource_size(res);
314         return size;
315 }
316
317 static bool __nd_namespace_blk_validate(struct nd_namespace_blk *nsblk)
318 {
319         struct nd_region *nd_region = to_nd_region(nsblk->common.dev.parent);
320         struct nd_mapping *nd_mapping = &nd_region->mapping[0];
321         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
322         struct nd_label_id label_id;
323         struct resource *res;
324         int count, i;
325
326         if (!nsblk->uuid || !nsblk->lbasize || !ndd)
327                 return false;
328
329         count = 0;
330         nd_label_gen_id(&label_id, nsblk->uuid, NSLABEL_FLAG_LOCAL);
331         for_each_dpa_resource(ndd, res) {
332                 if (strcmp(res->name, label_id.id) != 0)
333                         continue;
334                 /*
335                  * Resources with unacknowledged adjustments indicate a
336                  * failure to update labels
337                  */
338                 if (res->flags & DPA_RESOURCE_ADJUSTED)
339                         return false;
340                 count++;
341         }
342
343         /* These values match after a successful label update */
344         if (count != nsblk->num_resources)
345                 return false;
346
347         for (i = 0; i < nsblk->num_resources; i++) {
348                 struct resource *found = NULL;
349
350                 for_each_dpa_resource(ndd, res)
351                         if (res == nsblk->res[i]) {
352                                 found = res;
353                                 break;
354                         }
355                 /* stale resource */
356                 if (!found)
357                         return false;
358         }
359
360         return true;
361 }
362
363 resource_size_t nd_namespace_blk_validate(struct nd_namespace_blk *nsblk)
364 {
365         resource_size_t size;
366
367         nvdimm_bus_lock(&nsblk->common.dev);
368         size = __nd_namespace_blk_validate(nsblk);
369         nvdimm_bus_unlock(&nsblk->common.dev);
370
371         return size;
372 }
373 EXPORT_SYMBOL(nd_namespace_blk_validate);
374
375
376 static int nd_namespace_label_update(struct nd_region *nd_region,
377                 struct device *dev)
378 {
379         dev_WARN_ONCE(dev, dev->driver || to_ndns(dev)->claim,
380                         "namespace must be idle during label update\n");
381         if (dev->driver || to_ndns(dev)->claim)
382                 return 0;
383
384         /*
385          * Only allow label writes that will result in a valid namespace
386          * or deletion of an existing namespace.
387          */
388         if (is_namespace_pmem(dev)) {
389                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
390                 resource_size_t size = resource_size(&nspm->nsio.res);
391
392                 if (size == 0 && nspm->uuid)
393                         /* delete allocation */;
394                 else if (!nspm->uuid)
395                         return 0;
396
397                 return nd_pmem_namespace_label_update(nd_region, nspm, size);
398         } else if (is_namespace_blk(dev)) {
399                 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
400                 resource_size_t size = nd_namespace_blk_size(nsblk);
401
402                 if (size == 0 && nsblk->uuid)
403                         /* delete allocation */;
404                 else if (!nsblk->uuid || !nsblk->lbasize)
405                         return 0;
406
407                 return nd_blk_namespace_label_update(nd_region, nsblk, size);
408         } else
409                 return -ENXIO;
410 }
411
412 static ssize_t alt_name_store(struct device *dev,
413                 struct device_attribute *attr, const char *buf, size_t len)
414 {
415         struct nd_region *nd_region = to_nd_region(dev->parent);
416         ssize_t rc;
417
418         device_lock(dev);
419         nvdimm_bus_lock(dev);
420         wait_nvdimm_bus_probe_idle(dev);
421         rc = __alt_name_store(dev, buf, len);
422         if (rc >= 0)
423                 rc = nd_namespace_label_update(nd_region, dev);
424         dev_dbg(dev, "%s(%zd)\n", rc < 0 ? "fail " : "", rc);
425         nvdimm_bus_unlock(dev);
426         device_unlock(dev);
427
428         return rc < 0 ? rc : len;
429 }
430
431 static ssize_t alt_name_show(struct device *dev,
432                 struct device_attribute *attr, char *buf)
433 {
434         char *ns_altname;
435
436         if (is_namespace_pmem(dev)) {
437                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
438
439                 ns_altname = nspm->alt_name;
440         } else if (is_namespace_blk(dev)) {
441                 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
442
443                 ns_altname = nsblk->alt_name;
444         } else
445                 return -ENXIO;
446
447         return sprintf(buf, "%s\n", ns_altname ? ns_altname : "");
448 }
449 static DEVICE_ATTR_RW(alt_name);
450
451 static int scan_free(struct nd_region *nd_region,
452                 struct nd_mapping *nd_mapping, struct nd_label_id *label_id,
453                 resource_size_t n)
454 {
455         bool is_blk = strncmp(label_id->id, "blk", 3) == 0;
456         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
457         int rc = 0;
458
459         while (n) {
460                 struct resource *res, *last;
461                 resource_size_t new_start;
462
463                 last = NULL;
464                 for_each_dpa_resource(ndd, res)
465                         if (strcmp(res->name, label_id->id) == 0)
466                                 last = res;
467                 res = last;
468                 if (!res)
469                         return 0;
470
471                 if (n >= resource_size(res)) {
472                         n -= resource_size(res);
473                         nd_dbg_dpa(nd_region, ndd, res, "delete %d\n", rc);
474                         nvdimm_free_dpa(ndd, res);
475                         /* retry with last resource deleted */
476                         continue;
477                 }
478
479                 /*
480                  * Keep BLK allocations relegated to high DPA as much as
481                  * possible
482                  */
483                 if (is_blk)
484                         new_start = res->start + n;
485                 else
486                         new_start = res->start;
487
488                 rc = adjust_resource(res, new_start, resource_size(res) - n);
489                 if (rc == 0)
490                         res->flags |= DPA_RESOURCE_ADJUSTED;
491                 nd_dbg_dpa(nd_region, ndd, res, "shrink %d\n", rc);
492                 break;
493         }
494
495         return rc;
496 }
497
498 /**
499  * shrink_dpa_allocation - for each dimm in region free n bytes for label_id
500  * @nd_region: the set of dimms to reclaim @n bytes from
501  * @label_id: unique identifier for the namespace consuming this dpa range
502  * @n: number of bytes per-dimm to release
503  *
504  * Assumes resources are ordered.  Starting from the end try to
505  * adjust_resource() the allocation to @n, but if @n is larger than the
506  * allocation delete it and find the 'new' last allocation in the label
507  * set.
508  */
509 static int shrink_dpa_allocation(struct nd_region *nd_region,
510                 struct nd_label_id *label_id, resource_size_t n)
511 {
512         int i;
513
514         for (i = 0; i < nd_region->ndr_mappings; i++) {
515                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
516                 int rc;
517
518                 rc = scan_free(nd_region, nd_mapping, label_id, n);
519                 if (rc)
520                         return rc;
521         }
522
523         return 0;
524 }
525
526 static resource_size_t init_dpa_allocation(struct nd_label_id *label_id,
527                 struct nd_region *nd_region, struct nd_mapping *nd_mapping,
528                 resource_size_t n)
529 {
530         bool is_blk = strncmp(label_id->id, "blk", 3) == 0;
531         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
532         resource_size_t first_dpa;
533         struct resource *res;
534         int rc = 0;
535
536         /* allocate blk from highest dpa first */
537         if (is_blk)
538                 first_dpa = nd_mapping->start + nd_mapping->size - n;
539         else
540                 first_dpa = nd_mapping->start;
541
542         /* first resource allocation for this label-id or dimm */
543         res = nvdimm_allocate_dpa(ndd, label_id, first_dpa, n);
544         if (!res)
545                 rc = -EBUSY;
546
547         nd_dbg_dpa(nd_region, ndd, res, "init %d\n", rc);
548         return rc ? n : 0;
549 }
550
551
552 /**
553  * space_valid() - validate free dpa space against constraints
554  * @nd_region: hosting region of the free space
555  * @ndd: dimm device data for debug
556  * @label_id: namespace id to allocate space
557  * @prev: potential allocation that precedes free space
558  * @next: allocation that follows the given free space range
559  * @exist: first allocation with same id in the mapping
560  * @n: range that must satisfied for pmem allocations
561  * @valid: free space range to validate
562  *
563  * BLK-space is valid as long as it does not precede a PMEM
564  * allocation in a given region. PMEM-space must be contiguous
565  * and adjacent to an existing existing allocation (if one
566  * exists).  If reserving PMEM any space is valid.
567  */
568 static void space_valid(struct nd_region *nd_region, struct nvdimm_drvdata *ndd,
569                 struct nd_label_id *label_id, struct resource *prev,
570                 struct resource *next, struct resource *exist,
571                 resource_size_t n, struct resource *valid)
572 {
573         bool is_reserve = strcmp(label_id->id, "pmem-reserve") == 0;
574         bool is_pmem = strncmp(label_id->id, "pmem", 4) == 0;
575
576         if (valid->start >= valid->end)
577                 goto invalid;
578
579         if (is_reserve)
580                 return;
581
582         if (!is_pmem) {
583                 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
584                 struct nvdimm_bus *nvdimm_bus;
585                 struct blk_alloc_info info = {
586                         .nd_mapping = nd_mapping,
587                         .available = nd_mapping->size,
588                         .res = valid,
589                 };
590
591                 WARN_ON(!is_nd_blk(&nd_region->dev));
592                 nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev);
593                 device_for_each_child(&nvdimm_bus->dev, &info, alias_dpa_busy);
594                 return;
595         }
596
597         /* allocation needs to be contiguous, so this is all or nothing */
598         if (resource_size(valid) < n)
599                 goto invalid;
600
601         /* we've got all the space we need and no existing allocation */
602         if (!exist)
603                 return;
604
605         /* allocation needs to be contiguous with the existing namespace */
606         if (valid->start == exist->end + 1
607                         || valid->end == exist->start - 1)
608                 return;
609
610  invalid:
611         /* truncate @valid size to 0 */
612         valid->end = valid->start - 1;
613 }
614
615 enum alloc_loc {
616         ALLOC_ERR = 0, ALLOC_BEFORE, ALLOC_MID, ALLOC_AFTER,
617 };
618
619 static resource_size_t scan_allocate(struct nd_region *nd_region,
620                 struct nd_mapping *nd_mapping, struct nd_label_id *label_id,
621                 resource_size_t n)
622 {
623         resource_size_t mapping_end = nd_mapping->start + nd_mapping->size - 1;
624         bool is_pmem = strncmp(label_id->id, "pmem", 4) == 0;
625         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
626         struct resource *res, *exist = NULL, valid;
627         const resource_size_t to_allocate = n;
628         int first;
629
630         for_each_dpa_resource(ndd, res)
631                 if (strcmp(label_id->id, res->name) == 0)
632                         exist = res;
633
634         valid.start = nd_mapping->start;
635         valid.end = mapping_end;
636         valid.name = "free space";
637  retry:
638         first = 0;
639         for_each_dpa_resource(ndd, res) {
640                 struct resource *next = res->sibling, *new_res = NULL;
641                 resource_size_t allocate, available = 0;
642                 enum alloc_loc loc = ALLOC_ERR;
643                 const char *action;
644                 int rc = 0;
645
646                 /* ignore resources outside this nd_mapping */
647                 if (res->start > mapping_end)
648                         continue;
649                 if (res->end < nd_mapping->start)
650                         continue;
651
652                 /* space at the beginning of the mapping */
653                 if (!first++ && res->start > nd_mapping->start) {
654                         valid.start = nd_mapping->start;
655                         valid.end = res->start - 1;
656                         space_valid(nd_region, ndd, label_id, NULL, next, exist,
657                                         to_allocate, &valid);
658                         available = resource_size(&valid);
659                         if (available)
660                                 loc = ALLOC_BEFORE;
661                 }
662
663                 /* space between allocations */
664                 if (!loc && next) {
665                         valid.start = res->start + resource_size(res);
666                         valid.end = min(mapping_end, next->start - 1);
667                         space_valid(nd_region, ndd, label_id, res, next, exist,
668                                         to_allocate, &valid);
669                         available = resource_size(&valid);
670                         if (available)
671                                 loc = ALLOC_MID;
672                 }
673
674                 /* space at the end of the mapping */
675                 if (!loc && !next) {
676                         valid.start = res->start + resource_size(res);
677                         valid.end = mapping_end;
678                         space_valid(nd_region, ndd, label_id, res, next, exist,
679                                         to_allocate, &valid);
680                         available = resource_size(&valid);
681                         if (available)
682                                 loc = ALLOC_AFTER;
683                 }
684
685                 if (!loc || !available)
686                         continue;
687                 allocate = min(available, n);
688                 switch (loc) {
689                 case ALLOC_BEFORE:
690                         if (strcmp(res->name, label_id->id) == 0) {
691                                 /* adjust current resource up */
692                                 rc = adjust_resource(res, res->start - allocate,
693                                                 resource_size(res) + allocate);
694                                 action = "cur grow up";
695                         } else
696                                 action = "allocate";
697                         break;
698                 case ALLOC_MID:
699                         if (strcmp(next->name, label_id->id) == 0) {
700                                 /* adjust next resource up */
701                                 rc = adjust_resource(next, next->start
702                                                 - allocate, resource_size(next)
703                                                 + allocate);
704                                 new_res = next;
705                                 action = "next grow up";
706                         } else if (strcmp(res->name, label_id->id) == 0) {
707                                 action = "grow down";
708                         } else
709                                 action = "allocate";
710                         break;
711                 case ALLOC_AFTER:
712                         if (strcmp(res->name, label_id->id) == 0)
713                                 action = "grow down";
714                         else
715                                 action = "allocate";
716                         break;
717                 default:
718                         return n;
719                 }
720
721                 if (strcmp(action, "allocate") == 0) {
722                         /* BLK allocate bottom up */
723                         if (!is_pmem)
724                                 valid.start += available - allocate;
725
726                         new_res = nvdimm_allocate_dpa(ndd, label_id,
727                                         valid.start, allocate);
728                         if (!new_res)
729                                 rc = -EBUSY;
730                 } else if (strcmp(action, "grow down") == 0) {
731                         /* adjust current resource down */
732                         rc = adjust_resource(res, res->start, resource_size(res)
733                                         + allocate);
734                         if (rc == 0)
735                                 res->flags |= DPA_RESOURCE_ADJUSTED;
736                 }
737
738                 if (!new_res)
739                         new_res = res;
740
741                 nd_dbg_dpa(nd_region, ndd, new_res, "%s(%d) %d\n",
742                                 action, loc, rc);
743
744                 if (rc)
745                         return n;
746
747                 n -= allocate;
748                 if (n) {
749                         /*
750                          * Retry scan with newly inserted resources.
751                          * For example, if we did an ALLOC_BEFORE
752                          * insertion there may also have been space
753                          * available for an ALLOC_AFTER insertion, so we
754                          * need to check this same resource again
755                          */
756                         goto retry;
757                 } else
758                         return 0;
759         }
760
761         /*
762          * If we allocated nothing in the BLK case it may be because we are in
763          * an initial "pmem-reserve pass".  Only do an initial BLK allocation
764          * when none of the DPA space is reserved.
765          */
766         if ((is_pmem || !ndd->dpa.child) && n == to_allocate)
767                 return init_dpa_allocation(label_id, nd_region, nd_mapping, n);
768         return n;
769 }
770
771 static int merge_dpa(struct nd_region *nd_region,
772                 struct nd_mapping *nd_mapping, struct nd_label_id *label_id)
773 {
774         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
775         struct resource *res;
776
777         if (strncmp("pmem", label_id->id, 4) == 0)
778                 return 0;
779  retry:
780         for_each_dpa_resource(ndd, res) {
781                 int rc;
782                 struct resource *next = res->sibling;
783                 resource_size_t end = res->start + resource_size(res);
784
785                 if (!next || strcmp(res->name, label_id->id) != 0
786                                 || strcmp(next->name, label_id->id) != 0
787                                 || end != next->start)
788                         continue;
789                 end += resource_size(next);
790                 nvdimm_free_dpa(ndd, next);
791                 rc = adjust_resource(res, res->start, end - res->start);
792                 nd_dbg_dpa(nd_region, ndd, res, "merge %d\n", rc);
793                 if (rc)
794                         return rc;
795                 res->flags |= DPA_RESOURCE_ADJUSTED;
796                 goto retry;
797         }
798
799         return 0;
800 }
801
802 int __reserve_free_pmem(struct device *dev, void *data)
803 {
804         struct nvdimm *nvdimm = data;
805         struct nd_region *nd_region;
806         struct nd_label_id label_id;
807         int i;
808
809         if (!is_memory(dev))
810                 return 0;
811
812         nd_region = to_nd_region(dev);
813         if (nd_region->ndr_mappings == 0)
814                 return 0;
815
816         memset(&label_id, 0, sizeof(label_id));
817         strcat(label_id.id, "pmem-reserve");
818         for (i = 0; i < nd_region->ndr_mappings; i++) {
819                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
820                 resource_size_t n, rem = 0;
821
822                 if (nd_mapping->nvdimm != nvdimm)
823                         continue;
824
825                 n = nd_pmem_available_dpa(nd_region, nd_mapping, &rem);
826                 if (n == 0)
827                         return 0;
828                 rem = scan_allocate(nd_region, nd_mapping, &label_id, n);
829                 dev_WARN_ONCE(&nd_region->dev, rem,
830                                 "pmem reserve underrun: %#llx of %#llx bytes\n",
831                                 (unsigned long long) n - rem,
832                                 (unsigned long long) n);
833                 return rem ? -ENXIO : 0;
834         }
835
836         return 0;
837 }
838
839 void release_free_pmem(struct nvdimm_bus *nvdimm_bus,
840                 struct nd_mapping *nd_mapping)
841 {
842         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
843         struct resource *res, *_res;
844
845         for_each_dpa_resource_safe(ndd, res, _res)
846                 if (strcmp(res->name, "pmem-reserve") == 0)
847                         nvdimm_free_dpa(ndd, res);
848 }
849
850 static int reserve_free_pmem(struct nvdimm_bus *nvdimm_bus,
851                 struct nd_mapping *nd_mapping)
852 {
853         struct nvdimm *nvdimm = nd_mapping->nvdimm;
854         int rc;
855
856         rc = device_for_each_child(&nvdimm_bus->dev, nvdimm,
857                         __reserve_free_pmem);
858         if (rc)
859                 release_free_pmem(nvdimm_bus, nd_mapping);
860         return rc;
861 }
862
863 /**
864  * grow_dpa_allocation - for each dimm allocate n bytes for @label_id
865  * @nd_region: the set of dimms to allocate @n more bytes from
866  * @label_id: unique identifier for the namespace consuming this dpa range
867  * @n: number of bytes per-dimm to add to the existing allocation
868  *
869  * Assumes resources are ordered.  For BLK regions, first consume
870  * BLK-only available DPA free space, then consume PMEM-aliased DPA
871  * space starting at the highest DPA.  For PMEM regions start
872  * allocations from the start of an interleave set and end at the first
873  * BLK allocation or the end of the interleave set, whichever comes
874  * first.
875  */
876 static int grow_dpa_allocation(struct nd_region *nd_region,
877                 struct nd_label_id *label_id, resource_size_t n)
878 {
879         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev);
880         bool is_pmem = strncmp(label_id->id, "pmem", 4) == 0;
881         int i;
882
883         for (i = 0; i < nd_region->ndr_mappings; i++) {
884                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
885                 resource_size_t rem = n;
886                 int rc, j;
887
888                 /*
889                  * In the BLK case try once with all unallocated PMEM
890                  * reserved, and once without
891                  */
892                 for (j = is_pmem; j < 2; j++) {
893                         bool blk_only = j == 0;
894
895                         if (blk_only) {
896                                 rc = reserve_free_pmem(nvdimm_bus, nd_mapping);
897                                 if (rc)
898                                         return rc;
899                         }
900                         rem = scan_allocate(nd_region, nd_mapping,
901                                         label_id, rem);
902                         if (blk_only)
903                                 release_free_pmem(nvdimm_bus, nd_mapping);
904
905                         /* try again and allow encroachments into PMEM */
906                         if (rem == 0)
907                                 break;
908                 }
909
910                 dev_WARN_ONCE(&nd_region->dev, rem,
911                                 "allocation underrun: %#llx of %#llx bytes\n",
912                                 (unsigned long long) n - rem,
913                                 (unsigned long long) n);
914                 if (rem)
915                         return -ENXIO;
916
917                 rc = merge_dpa(nd_region, nd_mapping, label_id);
918                 if (rc)
919                         return rc;
920         }
921
922         return 0;
923 }
924
925 static void nd_namespace_pmem_set_resource(struct nd_region *nd_region,
926                 struct nd_namespace_pmem *nspm, resource_size_t size)
927 {
928         struct resource *res = &nspm->nsio.res;
929         resource_size_t offset = 0;
930
931         if (size && !nspm->uuid) {
932                 WARN_ON_ONCE(1);
933                 size = 0;
934         }
935
936         if (size && nspm->uuid) {
937                 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
938                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
939                 struct nd_label_id label_id;
940                 struct resource *res;
941
942                 if (!ndd) {
943                         size = 0;
944                         goto out;
945                 }
946
947                 nd_label_gen_id(&label_id, nspm->uuid, 0);
948
949                 /* calculate a spa offset from the dpa allocation offset */
950                 for_each_dpa_resource(ndd, res)
951                         if (strcmp(res->name, label_id.id) == 0) {
952                                 offset = (res->start - nd_mapping->start)
953                                         * nd_region->ndr_mappings;
954                                 goto out;
955                         }
956
957                 WARN_ON_ONCE(1);
958                 size = 0;
959         }
960
961  out:
962         res->start = nd_region->ndr_start + offset;
963         res->end = res->start + size - 1;
964 }
965
966 static bool uuid_not_set(const u8 *uuid, struct device *dev, const char *where)
967 {
968         if (!uuid) {
969                 dev_dbg(dev, "%s: uuid not set\n", where);
970                 return true;
971         }
972         return false;
973 }
974
975 static ssize_t __size_store(struct device *dev, unsigned long long val)
976 {
977         resource_size_t allocated = 0, available = 0;
978         struct nd_region *nd_region = to_nd_region(dev->parent);
979         struct nd_namespace_common *ndns = to_ndns(dev);
980         struct nd_mapping *nd_mapping;
981         struct nvdimm_drvdata *ndd;
982         struct nd_label_id label_id;
983         u32 flags = 0, remainder;
984         int rc, i, id = -1;
985         u8 *uuid = NULL;
986
987         if (dev->driver || ndns->claim)
988                 return -EBUSY;
989
990         if (is_namespace_pmem(dev)) {
991                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
992
993                 uuid = nspm->uuid;
994                 id = nspm->id;
995         } else if (is_namespace_blk(dev)) {
996                 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
997
998                 uuid = nsblk->uuid;
999                 flags = NSLABEL_FLAG_LOCAL;
1000                 id = nsblk->id;
1001         }
1002
1003         /*
1004          * We need a uuid for the allocation-label and dimm(s) on which
1005          * to store the label.
1006          */
1007         if (uuid_not_set(uuid, dev, __func__))
1008                 return -ENXIO;
1009         if (nd_region->ndr_mappings == 0) {
1010                 dev_dbg(dev, "not associated with dimm(s)\n");
1011                 return -ENXIO;
1012         }
1013
1014         div_u64_rem(val, SZ_4K * nd_region->ndr_mappings, &remainder);
1015         if (remainder) {
1016                 dev_dbg(dev, "%llu is not %dK aligned\n", val,
1017                                 (SZ_4K * nd_region->ndr_mappings) / SZ_1K);
1018                 return -EINVAL;
1019         }
1020
1021         nd_label_gen_id(&label_id, uuid, flags);
1022         for (i = 0; i < nd_region->ndr_mappings; i++) {
1023                 nd_mapping = &nd_region->mapping[i];
1024                 ndd = to_ndd(nd_mapping);
1025
1026                 /*
1027                  * All dimms in an interleave set, or the base dimm for a blk
1028                  * region, need to be enabled for the size to be changed.
1029                  */
1030                 if (!ndd)
1031                         return -ENXIO;
1032
1033                 allocated += nvdimm_allocated_dpa(ndd, &label_id);
1034         }
1035         available = nd_region_allocatable_dpa(nd_region);
1036
1037         if (val > available + allocated)
1038                 return -ENOSPC;
1039
1040         if (val == allocated)
1041                 return 0;
1042
1043         val = div_u64(val, nd_region->ndr_mappings);
1044         allocated = div_u64(allocated, nd_region->ndr_mappings);
1045         if (val < allocated)
1046                 rc = shrink_dpa_allocation(nd_region, &label_id,
1047                                 allocated - val);
1048         else
1049                 rc = grow_dpa_allocation(nd_region, &label_id, val - allocated);
1050
1051         if (rc)
1052                 return rc;
1053
1054         if (is_namespace_pmem(dev)) {
1055                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1056
1057                 nd_namespace_pmem_set_resource(nd_region, nspm,
1058                                 val * nd_region->ndr_mappings);
1059         }
1060
1061         /*
1062          * Try to delete the namespace if we deleted all of its
1063          * allocation, this is not the seed or 0th device for the
1064          * region, and it is not actively claimed by a btt, pfn, or dax
1065          * instance.
1066          */
1067         if (val == 0 && id != 0 && nd_region->ns_seed != dev && !ndns->claim)
1068                 nd_device_unregister(dev, ND_ASYNC);
1069
1070         return rc;
1071 }
1072
1073 static ssize_t size_store(struct device *dev,
1074                 struct device_attribute *attr, const char *buf, size_t len)
1075 {
1076         struct nd_region *nd_region = to_nd_region(dev->parent);
1077         unsigned long long val;
1078         u8 **uuid = NULL;
1079         int rc;
1080
1081         rc = kstrtoull(buf, 0, &val);
1082         if (rc)
1083                 return rc;
1084
1085         device_lock(dev);
1086         nvdimm_bus_lock(dev);
1087         wait_nvdimm_bus_probe_idle(dev);
1088         rc = __size_store(dev, val);
1089         if (rc >= 0)
1090                 rc = nd_namespace_label_update(nd_region, dev);
1091
1092         if (is_namespace_pmem(dev)) {
1093                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1094
1095                 uuid = &nspm->uuid;
1096         } else if (is_namespace_blk(dev)) {
1097                 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1098
1099                 uuid = &nsblk->uuid;
1100         }
1101
1102         if (rc == 0 && val == 0 && uuid) {
1103                 /* setting size zero == 'delete namespace' */
1104                 kfree(*uuid);
1105                 *uuid = NULL;
1106         }
1107
1108         dev_dbg(dev, "%llx %s (%d)\n", val, rc < 0 ? "fail" : "success", rc);
1109
1110         nvdimm_bus_unlock(dev);
1111         device_unlock(dev);
1112
1113         return rc < 0 ? rc : len;
1114 }
1115
1116 resource_size_t __nvdimm_namespace_capacity(struct nd_namespace_common *ndns)
1117 {
1118         struct device *dev = &ndns->dev;
1119
1120         if (is_namespace_pmem(dev)) {
1121                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1122
1123                 return resource_size(&nspm->nsio.res);
1124         } else if (is_namespace_blk(dev)) {
1125                 return nd_namespace_blk_size(to_nd_namespace_blk(dev));
1126         } else if (is_namespace_io(dev)) {
1127                 struct nd_namespace_io *nsio = to_nd_namespace_io(dev);
1128
1129                 return resource_size(&nsio->res);
1130         } else
1131                 WARN_ONCE(1, "unknown namespace type\n");
1132         return 0;
1133 }
1134
1135 resource_size_t nvdimm_namespace_capacity(struct nd_namespace_common *ndns)
1136 {
1137         resource_size_t size;
1138
1139         nvdimm_bus_lock(&ndns->dev);
1140         size = __nvdimm_namespace_capacity(ndns);
1141         nvdimm_bus_unlock(&ndns->dev);
1142
1143         return size;
1144 }
1145 EXPORT_SYMBOL(nvdimm_namespace_capacity);
1146
1147 bool nvdimm_namespace_locked(struct nd_namespace_common *ndns)
1148 {
1149         int i;
1150         bool locked = false;
1151         struct device *dev = &ndns->dev;
1152         struct nd_region *nd_region = to_nd_region(dev->parent);
1153
1154         for (i = 0; i < nd_region->ndr_mappings; i++) {
1155                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1156                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
1157
1158                 if (test_bit(NDD_LOCKED, &nvdimm->flags)) {
1159                         dev_dbg(dev, "%s locked\n", nvdimm_name(nvdimm));
1160                         locked = true;
1161                 }
1162         }
1163         return locked;
1164 }
1165 EXPORT_SYMBOL(nvdimm_namespace_locked);
1166
1167 static ssize_t size_show(struct device *dev,
1168                 struct device_attribute *attr, char *buf)
1169 {
1170         return sprintf(buf, "%llu\n", (unsigned long long)
1171                         nvdimm_namespace_capacity(to_ndns(dev)));
1172 }
1173 static DEVICE_ATTR(size, 0444, size_show, size_store);
1174
1175 static u8 *namespace_to_uuid(struct device *dev)
1176 {
1177         if (is_namespace_pmem(dev)) {
1178                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1179
1180                 return nspm->uuid;
1181         } else if (is_namespace_blk(dev)) {
1182                 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1183
1184                 return nsblk->uuid;
1185         } else
1186                 return ERR_PTR(-ENXIO);
1187 }
1188
1189 static ssize_t uuid_show(struct device *dev,
1190                 struct device_attribute *attr, char *buf)
1191 {
1192         u8 *uuid = namespace_to_uuid(dev);
1193
1194         if (IS_ERR(uuid))
1195                 return PTR_ERR(uuid);
1196         if (uuid)
1197                 return sprintf(buf, "%pUb\n", uuid);
1198         return sprintf(buf, "\n");
1199 }
1200
1201 /**
1202  * namespace_update_uuid - check for a unique uuid and whether we're "renaming"
1203  * @nd_region: parent region so we can updates all dimms in the set
1204  * @dev: namespace type for generating label_id
1205  * @new_uuid: incoming uuid
1206  * @old_uuid: reference to the uuid storage location in the namespace object
1207  */
1208 static int namespace_update_uuid(struct nd_region *nd_region,
1209                 struct device *dev, u8 *new_uuid, u8 **old_uuid)
1210 {
1211         u32 flags = is_namespace_blk(dev) ? NSLABEL_FLAG_LOCAL : 0;
1212         struct nd_label_id old_label_id;
1213         struct nd_label_id new_label_id;
1214         int i;
1215
1216         if (!nd_is_uuid_unique(dev, new_uuid))
1217                 return -EINVAL;
1218
1219         if (*old_uuid == NULL)
1220                 goto out;
1221
1222         /*
1223          * If we've already written a label with this uuid, then it's
1224          * too late to rename because we can't reliably update the uuid
1225          * without losing the old namespace.  Userspace must delete this
1226          * namespace to abandon the old uuid.
1227          */
1228         for (i = 0; i < nd_region->ndr_mappings; i++) {
1229                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1230
1231                 /*
1232                  * This check by itself is sufficient because old_uuid
1233                  * would be NULL above if this uuid did not exist in the
1234                  * currently written set.
1235                  *
1236                  * FIXME: can we delete uuid with zero dpa allocated?
1237                  */
1238                 if (list_empty(&nd_mapping->labels))
1239                         return -EBUSY;
1240         }
1241
1242         nd_label_gen_id(&old_label_id, *old_uuid, flags);
1243         nd_label_gen_id(&new_label_id, new_uuid, flags);
1244         for (i = 0; i < nd_region->ndr_mappings; i++) {
1245                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1246                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1247                 struct resource *res;
1248
1249                 for_each_dpa_resource(ndd, res)
1250                         if (strcmp(res->name, old_label_id.id) == 0)
1251                                 sprintf((void *) res->name, "%s",
1252                                                 new_label_id.id);
1253         }
1254         kfree(*old_uuid);
1255  out:
1256         *old_uuid = new_uuid;
1257         return 0;
1258 }
1259
1260 static ssize_t uuid_store(struct device *dev,
1261                 struct device_attribute *attr, const char *buf, size_t len)
1262 {
1263         struct nd_region *nd_region = to_nd_region(dev->parent);
1264         u8 *uuid = NULL;
1265         ssize_t rc = 0;
1266         u8 **ns_uuid;
1267
1268         if (is_namespace_pmem(dev)) {
1269                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1270
1271                 ns_uuid = &nspm->uuid;
1272         } else if (is_namespace_blk(dev)) {
1273                 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1274
1275                 ns_uuid = &nsblk->uuid;
1276         } else
1277                 return -ENXIO;
1278
1279         device_lock(dev);
1280         nvdimm_bus_lock(dev);
1281         wait_nvdimm_bus_probe_idle(dev);
1282         if (to_ndns(dev)->claim)
1283                 rc = -EBUSY;
1284         if (rc >= 0)
1285                 rc = nd_uuid_store(dev, &uuid, buf, len);
1286         if (rc >= 0)
1287                 rc = namespace_update_uuid(nd_region, dev, uuid, ns_uuid);
1288         if (rc >= 0)
1289                 rc = nd_namespace_label_update(nd_region, dev);
1290         else
1291                 kfree(uuid);
1292         dev_dbg(dev, "result: %zd wrote: %s%s", rc, buf,
1293                         buf[len - 1] == '\n' ? "" : "\n");
1294         nvdimm_bus_unlock(dev);
1295         device_unlock(dev);
1296
1297         return rc < 0 ? rc : len;
1298 }
1299 static DEVICE_ATTR_RW(uuid);
1300
1301 static ssize_t resource_show(struct device *dev,
1302                 struct device_attribute *attr, char *buf)
1303 {
1304         struct resource *res;
1305
1306         if (is_namespace_pmem(dev)) {
1307                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1308
1309                 res = &nspm->nsio.res;
1310         } else if (is_namespace_io(dev)) {
1311                 struct nd_namespace_io *nsio = to_nd_namespace_io(dev);
1312
1313                 res = &nsio->res;
1314         } else
1315                 return -ENXIO;
1316
1317         /* no address to convey if the namespace has no allocation */
1318         if (resource_size(res) == 0)
1319                 return -ENXIO;
1320         return sprintf(buf, "%#llx\n", (unsigned long long) res->start);
1321 }
1322 static DEVICE_ATTR_RO(resource);
1323
1324 static const unsigned long blk_lbasize_supported[] = { 512, 520, 528,
1325         4096, 4104, 4160, 4224, 0 };
1326
1327 static const unsigned long pmem_lbasize_supported[] = { 512, 4096, 0 };
1328
1329 static ssize_t sector_size_show(struct device *dev,
1330                 struct device_attribute *attr, char *buf)
1331 {
1332         if (is_namespace_blk(dev)) {
1333                 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1334
1335                 return nd_size_select_show(nsblk->lbasize,
1336                                 blk_lbasize_supported, buf);
1337         }
1338
1339         if (is_namespace_pmem(dev)) {
1340                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1341
1342                 return nd_size_select_show(nspm->lbasize,
1343                                 pmem_lbasize_supported, buf);
1344         }
1345         return -ENXIO;
1346 }
1347
1348 static ssize_t sector_size_store(struct device *dev,
1349                 struct device_attribute *attr, const char *buf, size_t len)
1350 {
1351         struct nd_region *nd_region = to_nd_region(dev->parent);
1352         const unsigned long *supported;
1353         unsigned long *lbasize;
1354         ssize_t rc = 0;
1355
1356         if (is_namespace_blk(dev)) {
1357                 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1358
1359                 lbasize = &nsblk->lbasize;
1360                 supported = blk_lbasize_supported;
1361         } else if (is_namespace_pmem(dev)) {
1362                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1363
1364                 lbasize = &nspm->lbasize;
1365                 supported = pmem_lbasize_supported;
1366         } else
1367                 return -ENXIO;
1368
1369         device_lock(dev);
1370         nvdimm_bus_lock(dev);
1371         if (to_ndns(dev)->claim)
1372                 rc = -EBUSY;
1373         if (rc >= 0)
1374                 rc = nd_size_select_store(dev, buf, lbasize, supported);
1375         if (rc >= 0)
1376                 rc = nd_namespace_label_update(nd_region, dev);
1377         dev_dbg(dev, "result: %zd %s: %s%s", rc, rc < 0 ? "tried" : "wrote",
1378                         buf, buf[len - 1] == '\n' ? "" : "\n");
1379         nvdimm_bus_unlock(dev);
1380         device_unlock(dev);
1381
1382         return rc ? rc : len;
1383 }
1384 static DEVICE_ATTR_RW(sector_size);
1385
1386 static ssize_t dpa_extents_show(struct device *dev,
1387                 struct device_attribute *attr, char *buf)
1388 {
1389         struct nd_region *nd_region = to_nd_region(dev->parent);
1390         struct nd_label_id label_id;
1391         int count = 0, i;
1392         u8 *uuid = NULL;
1393         u32 flags = 0;
1394
1395         nvdimm_bus_lock(dev);
1396         if (is_namespace_pmem(dev)) {
1397                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1398
1399                 uuid = nspm->uuid;
1400                 flags = 0;
1401         } else if (is_namespace_blk(dev)) {
1402                 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1403
1404                 uuid = nsblk->uuid;
1405                 flags = NSLABEL_FLAG_LOCAL;
1406         }
1407
1408         if (!uuid)
1409                 goto out;
1410
1411         nd_label_gen_id(&label_id, uuid, flags);
1412         for (i = 0; i < nd_region->ndr_mappings; i++) {
1413                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1414                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1415                 struct resource *res;
1416
1417                 for_each_dpa_resource(ndd, res)
1418                         if (strcmp(res->name, label_id.id) == 0)
1419                                 count++;
1420         }
1421  out:
1422         nvdimm_bus_unlock(dev);
1423
1424         return sprintf(buf, "%d\n", count);
1425 }
1426 static DEVICE_ATTR_RO(dpa_extents);
1427
1428 static int btt_claim_class(struct device *dev)
1429 {
1430         struct nd_region *nd_region = to_nd_region(dev->parent);
1431         int i, loop_bitmask = 0;
1432
1433         for (i = 0; i < nd_region->ndr_mappings; i++) {
1434                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1435                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1436                 struct nd_namespace_index *nsindex;
1437
1438                 /*
1439                  * If any of the DIMMs do not support labels the only
1440                  * possible BTT format is v1.
1441                  */
1442                 if (!ndd) {
1443                         loop_bitmask = 0;
1444                         break;
1445                 }
1446
1447                 nsindex = to_namespace_index(ndd, ndd->ns_current);
1448                 if (nsindex == NULL)
1449                         loop_bitmask |= 1;
1450                 else {
1451                         /* check whether existing labels are v1.1 or v1.2 */
1452                         if (__le16_to_cpu(nsindex->major) == 1
1453                                         && __le16_to_cpu(nsindex->minor) == 1)
1454                                 loop_bitmask |= 2;
1455                         else
1456                                 loop_bitmask |= 4;
1457                 }
1458         }
1459         /*
1460          * If nsindex is null loop_bitmask's bit 0 will be set, and if an index
1461          * block is found, a v1.1 label for any mapping will set bit 1, and a
1462          * v1.2 label will set bit 2.
1463          *
1464          * At the end of the loop, at most one of the three bits must be set.
1465          * If multiple bits were set, it means the different mappings disagree
1466          * about their labels, and this must be cleaned up first.
1467          *
1468          * If all the label index blocks are found to agree, nsindex of NULL
1469          * implies labels haven't been initialized yet, and when they will,
1470          * they will be of the 1.2 format, so we can assume BTT2.0
1471          *
1472          * If 1.1 labels are found, we enforce BTT1.1, and if 1.2 labels are
1473          * found, we enforce BTT2.0
1474          *
1475          * If the loop was never entered, default to BTT1.1 (legacy namespaces)
1476          */
1477         switch (loop_bitmask) {
1478         case 0:
1479         case 2:
1480                 return NVDIMM_CCLASS_BTT;
1481         case 1:
1482         case 4:
1483                 return NVDIMM_CCLASS_BTT2;
1484         default:
1485                 return -ENXIO;
1486         }
1487 }
1488
1489 static ssize_t holder_show(struct device *dev,
1490                 struct device_attribute *attr, char *buf)
1491 {
1492         struct nd_namespace_common *ndns = to_ndns(dev);
1493         ssize_t rc;
1494
1495         device_lock(dev);
1496         rc = sprintf(buf, "%s\n", ndns->claim ? dev_name(ndns->claim) : "");
1497         device_unlock(dev);
1498
1499         return rc;
1500 }
1501 static DEVICE_ATTR_RO(holder);
1502
1503 static ssize_t __holder_class_store(struct device *dev, const char *buf)
1504 {
1505         struct nd_namespace_common *ndns = to_ndns(dev);
1506
1507         if (dev->driver || ndns->claim)
1508                 return -EBUSY;
1509
1510         if (strcmp(buf, "btt") == 0 || strcmp(buf, "btt\n") == 0)
1511                 ndns->claim_class = btt_claim_class(dev);
1512         else if (strcmp(buf, "pfn") == 0 || strcmp(buf, "pfn\n") == 0)
1513                 ndns->claim_class = NVDIMM_CCLASS_PFN;
1514         else if (strcmp(buf, "dax") == 0 || strcmp(buf, "dax\n") == 0)
1515                 ndns->claim_class = NVDIMM_CCLASS_DAX;
1516         else if (strcmp(buf, "") == 0 || strcmp(buf, "\n") == 0)
1517                 ndns->claim_class = NVDIMM_CCLASS_NONE;
1518         else
1519                 return -EINVAL;
1520
1521         /* btt_claim_class() could've returned an error */
1522         if (ndns->claim_class < 0)
1523                 return ndns->claim_class;
1524
1525         return 0;
1526 }
1527
1528 static ssize_t holder_class_store(struct device *dev,
1529                 struct device_attribute *attr, const char *buf, size_t len)
1530 {
1531         struct nd_region *nd_region = to_nd_region(dev->parent);
1532         ssize_t rc;
1533
1534         device_lock(dev);
1535         nvdimm_bus_lock(dev);
1536         wait_nvdimm_bus_probe_idle(dev);
1537         rc = __holder_class_store(dev, buf);
1538         if (rc >= 0)
1539                 rc = nd_namespace_label_update(nd_region, dev);
1540         dev_dbg(dev, "%s(%zd)\n", rc < 0 ? "fail " : "", rc);
1541         nvdimm_bus_unlock(dev);
1542         device_unlock(dev);
1543
1544         return rc < 0 ? rc : len;
1545 }
1546
1547 static ssize_t holder_class_show(struct device *dev,
1548                 struct device_attribute *attr, char *buf)
1549 {
1550         struct nd_namespace_common *ndns = to_ndns(dev);
1551         ssize_t rc;
1552
1553         device_lock(dev);
1554         if (ndns->claim_class == NVDIMM_CCLASS_NONE)
1555                 rc = sprintf(buf, "\n");
1556         else if ((ndns->claim_class == NVDIMM_CCLASS_BTT) ||
1557                         (ndns->claim_class == NVDIMM_CCLASS_BTT2))
1558                 rc = sprintf(buf, "btt\n");
1559         else if (ndns->claim_class == NVDIMM_CCLASS_PFN)
1560                 rc = sprintf(buf, "pfn\n");
1561         else if (ndns->claim_class == NVDIMM_CCLASS_DAX)
1562                 rc = sprintf(buf, "dax\n");
1563         else
1564                 rc = sprintf(buf, "<unknown>\n");
1565         device_unlock(dev);
1566
1567         return rc;
1568 }
1569 static DEVICE_ATTR_RW(holder_class);
1570
1571 static ssize_t mode_show(struct device *dev,
1572                 struct device_attribute *attr, char *buf)
1573 {
1574         struct nd_namespace_common *ndns = to_ndns(dev);
1575         struct device *claim;
1576         char *mode;
1577         ssize_t rc;
1578
1579         device_lock(dev);
1580         claim = ndns->claim;
1581         if (claim && is_nd_btt(claim))
1582                 mode = "safe";
1583         else if (claim && is_nd_pfn(claim))
1584                 mode = "memory";
1585         else if (claim && is_nd_dax(claim))
1586                 mode = "dax";
1587         else if (!claim && pmem_should_map_pages(dev))
1588                 mode = "memory";
1589         else
1590                 mode = "raw";
1591         rc = sprintf(buf, "%s\n", mode);
1592         device_unlock(dev);
1593
1594         return rc;
1595 }
1596 static DEVICE_ATTR_RO(mode);
1597
1598 static ssize_t force_raw_store(struct device *dev,
1599                 struct device_attribute *attr, const char *buf, size_t len)
1600 {
1601         bool force_raw;
1602         int rc = strtobool(buf, &force_raw);
1603
1604         if (rc)
1605                 return rc;
1606
1607         to_ndns(dev)->force_raw = force_raw;
1608         return len;
1609 }
1610
1611 static ssize_t force_raw_show(struct device *dev,
1612                 struct device_attribute *attr, char *buf)
1613 {
1614         return sprintf(buf, "%d\n", to_ndns(dev)->force_raw);
1615 }
1616 static DEVICE_ATTR_RW(force_raw);
1617
1618 static struct attribute *nd_namespace_attributes[] = {
1619         &dev_attr_nstype.attr,
1620         &dev_attr_size.attr,
1621         &dev_attr_mode.attr,
1622         &dev_attr_uuid.attr,
1623         &dev_attr_holder.attr,
1624         &dev_attr_resource.attr,
1625         &dev_attr_alt_name.attr,
1626         &dev_attr_force_raw.attr,
1627         &dev_attr_sector_size.attr,
1628         &dev_attr_dpa_extents.attr,
1629         &dev_attr_holder_class.attr,
1630         NULL,
1631 };
1632
1633 static umode_t namespace_visible(struct kobject *kobj,
1634                 struct attribute *a, int n)
1635 {
1636         struct device *dev = container_of(kobj, struct device, kobj);
1637
1638         if (a == &dev_attr_resource.attr) {
1639                 if (is_namespace_blk(dev))
1640                         return 0;
1641                 return 0400;
1642         }
1643
1644         if (is_namespace_pmem(dev) || is_namespace_blk(dev)) {
1645                 if (a == &dev_attr_size.attr)
1646                         return 0644;
1647
1648                 return a->mode;
1649         }
1650
1651         if (a == &dev_attr_nstype.attr || a == &dev_attr_size.attr
1652                         || a == &dev_attr_holder.attr
1653                         || a == &dev_attr_holder_class.attr
1654                         || a == &dev_attr_force_raw.attr
1655                         || a == &dev_attr_mode.attr)
1656                 return a->mode;
1657
1658         return 0;
1659 }
1660
1661 static struct attribute_group nd_namespace_attribute_group = {
1662         .attrs = nd_namespace_attributes,
1663         .is_visible = namespace_visible,
1664 };
1665
1666 static const struct attribute_group *nd_namespace_attribute_groups[] = {
1667         &nd_device_attribute_group,
1668         &nd_namespace_attribute_group,
1669         &nd_numa_attribute_group,
1670         NULL,
1671 };
1672
1673 struct nd_namespace_common *nvdimm_namespace_common_probe(struct device *dev)
1674 {
1675         struct nd_btt *nd_btt = is_nd_btt(dev) ? to_nd_btt(dev) : NULL;
1676         struct nd_pfn *nd_pfn = is_nd_pfn(dev) ? to_nd_pfn(dev) : NULL;
1677         struct nd_dax *nd_dax = is_nd_dax(dev) ? to_nd_dax(dev) : NULL;
1678         struct nd_namespace_common *ndns = NULL;
1679         resource_size_t size;
1680
1681         if (nd_btt || nd_pfn || nd_dax) {
1682                 if (nd_btt)
1683                         ndns = nd_btt->ndns;
1684                 else if (nd_pfn)
1685                         ndns = nd_pfn->ndns;
1686                 else if (nd_dax)
1687                         ndns = nd_dax->nd_pfn.ndns;
1688
1689                 if (!ndns)
1690                         return ERR_PTR(-ENODEV);
1691
1692                 /*
1693                  * Flush any in-progess probes / removals in the driver
1694                  * for the raw personality of this namespace.
1695                  */
1696                 device_lock(&ndns->dev);
1697                 device_unlock(&ndns->dev);
1698                 if (ndns->dev.driver) {
1699                         dev_dbg(&ndns->dev, "is active, can't bind %s\n",
1700                                         dev_name(dev));
1701                         return ERR_PTR(-EBUSY);
1702                 }
1703                 if (dev_WARN_ONCE(&ndns->dev, ndns->claim != dev,
1704                                         "host (%s) vs claim (%s) mismatch\n",
1705                                         dev_name(dev),
1706                                         dev_name(ndns->claim)))
1707                         return ERR_PTR(-ENXIO);
1708         } else {
1709                 ndns = to_ndns(dev);
1710                 if (ndns->claim) {
1711                         dev_dbg(dev, "claimed by %s, failing probe\n",
1712                                 dev_name(ndns->claim));
1713
1714                         return ERR_PTR(-ENXIO);
1715                 }
1716         }
1717
1718         if (nvdimm_namespace_locked(ndns))
1719                 return ERR_PTR(-EACCES);
1720
1721         size = nvdimm_namespace_capacity(ndns);
1722         if (size < ND_MIN_NAMESPACE_SIZE) {
1723                 dev_dbg(&ndns->dev, "%pa, too small must be at least %#x\n",
1724                                 &size, ND_MIN_NAMESPACE_SIZE);
1725                 return ERR_PTR(-ENODEV);
1726         }
1727
1728         if (is_namespace_pmem(&ndns->dev)) {
1729                 struct nd_namespace_pmem *nspm;
1730
1731                 nspm = to_nd_namespace_pmem(&ndns->dev);
1732                 if (uuid_not_set(nspm->uuid, &ndns->dev, __func__))
1733                         return ERR_PTR(-ENODEV);
1734         } else if (is_namespace_blk(&ndns->dev)) {
1735                 struct nd_namespace_blk *nsblk;
1736
1737                 nsblk = to_nd_namespace_blk(&ndns->dev);
1738                 if (uuid_not_set(nsblk->uuid, &ndns->dev, __func__))
1739                         return ERR_PTR(-ENODEV);
1740                 if (!nsblk->lbasize) {
1741                         dev_dbg(&ndns->dev, "sector size not set\n");
1742                         return ERR_PTR(-ENODEV);
1743                 }
1744                 if (!nd_namespace_blk_validate(nsblk))
1745                         return ERR_PTR(-ENODEV);
1746         }
1747
1748         return ndns;
1749 }
1750 EXPORT_SYMBOL(nvdimm_namespace_common_probe);
1751
1752 static struct device **create_namespace_io(struct nd_region *nd_region)
1753 {
1754         struct nd_namespace_io *nsio;
1755         struct device *dev, **devs;
1756         struct resource *res;
1757
1758         nsio = kzalloc(sizeof(*nsio), GFP_KERNEL);
1759         if (!nsio)
1760                 return NULL;
1761
1762         devs = kcalloc(2, sizeof(struct device *), GFP_KERNEL);
1763         if (!devs) {
1764                 kfree(nsio);
1765                 return NULL;
1766         }
1767
1768         dev = &nsio->common.dev;
1769         dev->type = &namespace_io_device_type;
1770         dev->parent = &nd_region->dev;
1771         res = &nsio->res;
1772         res->name = dev_name(&nd_region->dev);
1773         res->flags = IORESOURCE_MEM;
1774         res->start = nd_region->ndr_start;
1775         res->end = res->start + nd_region->ndr_size - 1;
1776
1777         devs[0] = dev;
1778         return devs;
1779 }
1780
1781 static bool has_uuid_at_pos(struct nd_region *nd_region, u8 *uuid,
1782                 u64 cookie, u16 pos)
1783 {
1784         struct nd_namespace_label *found = NULL;
1785         int i;
1786
1787         for (i = 0; i < nd_region->ndr_mappings; i++) {
1788                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1789                 struct nd_interleave_set *nd_set = nd_region->nd_set;
1790                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1791                 struct nd_label_ent *label_ent;
1792                 bool found_uuid = false;
1793
1794                 list_for_each_entry(label_ent, &nd_mapping->labels, list) {
1795                         struct nd_namespace_label *nd_label = label_ent->label;
1796                         u16 position, nlabel;
1797                         u64 isetcookie;
1798
1799                         if (!nd_label)
1800                                 continue;
1801                         isetcookie = __le64_to_cpu(nd_label->isetcookie);
1802                         position = __le16_to_cpu(nd_label->position);
1803                         nlabel = __le16_to_cpu(nd_label->nlabel);
1804
1805                         if (isetcookie != cookie)
1806                                 continue;
1807
1808                         if (memcmp(nd_label->uuid, uuid, NSLABEL_UUID_LEN) != 0)
1809                                 continue;
1810
1811                         if (namespace_label_has(ndd, type_guid)
1812                                         && !guid_equal(&nd_set->type_guid,
1813                                                 &nd_label->type_guid)) {
1814                                 dev_dbg(ndd->dev, "expect type_guid %pUb got %pUb\n",
1815                                                 nd_set->type_guid.b,
1816                                                 nd_label->type_guid.b);
1817                                 continue;
1818                         }
1819
1820                         if (found_uuid) {
1821                                 dev_dbg(ndd->dev, "duplicate entry for uuid\n");
1822                                 return false;
1823                         }
1824                         found_uuid = true;
1825                         if (nlabel != nd_region->ndr_mappings)
1826                                 continue;
1827                         if (position != pos)
1828                                 continue;
1829                         found = nd_label;
1830                         break;
1831                 }
1832                 if (found)
1833                         break;
1834         }
1835         return found != NULL;
1836 }
1837
1838 static int select_pmem_id(struct nd_region *nd_region, u8 *pmem_id)
1839 {
1840         int i;
1841
1842         if (!pmem_id)
1843                 return -ENODEV;
1844
1845         for (i = 0; i < nd_region->ndr_mappings; i++) {
1846                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1847                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1848                 struct nd_namespace_label *nd_label = NULL;
1849                 u64 hw_start, hw_end, pmem_start, pmem_end;
1850                 struct nd_label_ent *label_ent;
1851
1852                 lockdep_assert_held(&nd_mapping->lock);
1853                 list_for_each_entry(label_ent, &nd_mapping->labels, list) {
1854                         nd_label = label_ent->label;
1855                         if (!nd_label)
1856                                 continue;
1857                         if (memcmp(nd_label->uuid, pmem_id, NSLABEL_UUID_LEN) == 0)
1858                                 break;
1859                         nd_label = NULL;
1860                 }
1861
1862                 if (!nd_label) {
1863                         WARN_ON(1);
1864                         return -EINVAL;
1865                 }
1866
1867                 /*
1868                  * Check that this label is compliant with the dpa
1869                  * range published in NFIT
1870                  */
1871                 hw_start = nd_mapping->start;
1872                 hw_end = hw_start + nd_mapping->size;
1873                 pmem_start = __le64_to_cpu(nd_label->dpa);
1874                 pmem_end = pmem_start + __le64_to_cpu(nd_label->rawsize);
1875                 if (pmem_start >= hw_start && pmem_start < hw_end
1876                                 && pmem_end <= hw_end && pmem_end > hw_start)
1877                         /* pass */;
1878                 else {
1879                         dev_dbg(&nd_region->dev, "%s invalid label for %pUb\n",
1880                                         dev_name(ndd->dev), nd_label->uuid);
1881                         return -EINVAL;
1882                 }
1883
1884                 /* move recently validated label to the front of the list */
1885                 list_move(&label_ent->list, &nd_mapping->labels);
1886         }
1887         return 0;
1888 }
1889
1890 /**
1891  * create_namespace_pmem - validate interleave set labelling, retrieve label0
1892  * @nd_region: region with mappings to validate
1893  * @nspm: target namespace to create
1894  * @nd_label: target pmem namespace label to evaluate
1895  */
1896 static struct device *create_namespace_pmem(struct nd_region *nd_region,
1897                 struct nd_namespace_index *nsindex,
1898                 struct nd_namespace_label *nd_label)
1899 {
1900         u64 cookie = nd_region_interleave_set_cookie(nd_region, nsindex);
1901         u64 altcookie = nd_region_interleave_set_altcookie(nd_region);
1902         struct nd_label_ent *label_ent;
1903         struct nd_namespace_pmem *nspm;
1904         struct nd_mapping *nd_mapping;
1905         resource_size_t size = 0;
1906         struct resource *res;
1907         struct device *dev;
1908         int rc = 0;
1909         u16 i;
1910
1911         if (cookie == 0) {
1912                 dev_dbg(&nd_region->dev, "invalid interleave-set-cookie\n");
1913                 return ERR_PTR(-ENXIO);
1914         }
1915
1916         if (__le64_to_cpu(nd_label->isetcookie) != cookie) {
1917                 dev_dbg(&nd_region->dev, "invalid cookie in label: %pUb\n",
1918                                 nd_label->uuid);
1919                 if (__le64_to_cpu(nd_label->isetcookie) != altcookie)
1920                         return ERR_PTR(-EAGAIN);
1921
1922                 dev_dbg(&nd_region->dev, "valid altcookie in label: %pUb\n",
1923                                 nd_label->uuid);
1924         }
1925
1926         nspm = kzalloc(sizeof(*nspm), GFP_KERNEL);
1927         if (!nspm)
1928                 return ERR_PTR(-ENOMEM);
1929
1930         nspm->id = -1;
1931         dev = &nspm->nsio.common.dev;
1932         dev->type = &namespace_pmem_device_type;
1933         dev->parent = &nd_region->dev;
1934         res = &nspm->nsio.res;
1935         res->name = dev_name(&nd_region->dev);
1936         res->flags = IORESOURCE_MEM;
1937
1938         for (i = 0; i < nd_region->ndr_mappings; i++) {
1939                 if (has_uuid_at_pos(nd_region, nd_label->uuid, cookie, i))
1940                         continue;
1941                 if (has_uuid_at_pos(nd_region, nd_label->uuid, altcookie, i))
1942                         continue;
1943                 break;
1944         }
1945
1946         if (i < nd_region->ndr_mappings) {
1947                 struct nvdimm *nvdimm = nd_region->mapping[i].nvdimm;
1948
1949                 /*
1950                  * Give up if we don't find an instance of a uuid at each
1951                  * position (from 0 to nd_region->ndr_mappings - 1), or if we
1952                  * find a dimm with two instances of the same uuid.
1953                  */
1954                 dev_err(&nd_region->dev, "%s missing label for %pUb\n",
1955                                 nvdimm_name(nvdimm), nd_label->uuid);
1956                 rc = -EINVAL;
1957                 goto err;
1958         }
1959
1960         /*
1961          * Fix up each mapping's 'labels' to have the validated pmem label for
1962          * that position at labels[0], and NULL at labels[1].  In the process,
1963          * check that the namespace aligns with interleave-set.  We know
1964          * that it does not overlap with any blk namespaces by virtue of
1965          * the dimm being enabled (i.e. nd_label_reserve_dpa()
1966          * succeeded).
1967          */
1968         rc = select_pmem_id(nd_region, nd_label->uuid);
1969         if (rc)
1970                 goto err;
1971
1972         /* Calculate total size and populate namespace properties from label0 */
1973         for (i = 0; i < nd_region->ndr_mappings; i++) {
1974                 struct nd_namespace_label *label0;
1975                 struct nvdimm_drvdata *ndd;
1976
1977                 nd_mapping = &nd_region->mapping[i];
1978                 label_ent = list_first_entry_or_null(&nd_mapping->labels,
1979                                 typeof(*label_ent), list);
1980                 label0 = label_ent ? label_ent->label : 0;
1981
1982                 if (!label0) {
1983                         WARN_ON(1);
1984                         continue;
1985                 }
1986
1987                 size += __le64_to_cpu(label0->rawsize);
1988                 if (__le16_to_cpu(label0->position) != 0)
1989                         continue;
1990                 WARN_ON(nspm->alt_name || nspm->uuid);
1991                 nspm->alt_name = kmemdup((void __force *) label0->name,
1992                                 NSLABEL_NAME_LEN, GFP_KERNEL);
1993                 nspm->uuid = kmemdup((void __force *) label0->uuid,
1994                                 NSLABEL_UUID_LEN, GFP_KERNEL);
1995                 nspm->lbasize = __le64_to_cpu(label0->lbasize);
1996                 ndd = to_ndd(nd_mapping);
1997                 if (namespace_label_has(ndd, abstraction_guid))
1998                         nspm->nsio.common.claim_class
1999                                 = to_nvdimm_cclass(&label0->abstraction_guid);
2000
2001         }
2002
2003         if (!nspm->alt_name || !nspm->uuid) {
2004                 rc = -ENOMEM;
2005                 goto err;
2006         }
2007
2008         nd_namespace_pmem_set_resource(nd_region, nspm, size);
2009
2010         return dev;
2011  err:
2012         namespace_pmem_release(dev);
2013         switch (rc) {
2014         case -EINVAL:
2015                 dev_dbg(&nd_region->dev, "invalid label(s)\n");
2016                 break;
2017         case -ENODEV:
2018                 dev_dbg(&nd_region->dev, "label not found\n");
2019                 break;
2020         default:
2021                 dev_dbg(&nd_region->dev, "unexpected err: %d\n", rc);
2022                 break;
2023         }
2024         return ERR_PTR(rc);
2025 }
2026
2027 struct resource *nsblk_add_resource(struct nd_region *nd_region,
2028                 struct nvdimm_drvdata *ndd, struct nd_namespace_blk *nsblk,
2029                 resource_size_t start)
2030 {
2031         struct nd_label_id label_id;
2032         struct resource *res;
2033
2034         nd_label_gen_id(&label_id, nsblk->uuid, NSLABEL_FLAG_LOCAL);
2035         res = krealloc(nsblk->res,
2036                         sizeof(void *) * (nsblk->num_resources + 1),
2037                         GFP_KERNEL);
2038         if (!res)
2039                 return NULL;
2040         nsblk->res = (struct resource **) res;
2041         for_each_dpa_resource(ndd, res)
2042                 if (strcmp(res->name, label_id.id) == 0
2043                                 && res->start == start) {
2044                         nsblk->res[nsblk->num_resources++] = res;
2045                         return res;
2046                 }
2047         return NULL;
2048 }
2049
2050 static struct device *nd_namespace_blk_create(struct nd_region *nd_region)
2051 {
2052         struct nd_namespace_blk *nsblk;
2053         struct device *dev;
2054
2055         if (!is_nd_blk(&nd_region->dev))
2056                 return NULL;
2057
2058         nsblk = kzalloc(sizeof(*nsblk), GFP_KERNEL);
2059         if (!nsblk)
2060                 return NULL;
2061
2062         dev = &nsblk->common.dev;
2063         dev->type = &namespace_blk_device_type;
2064         nsblk->id = ida_simple_get(&nd_region->ns_ida, 0, 0, GFP_KERNEL);
2065         if (nsblk->id < 0) {
2066                 kfree(nsblk);
2067                 return NULL;
2068         }
2069         dev_set_name(dev, "namespace%d.%d", nd_region->id, nsblk->id);
2070         dev->parent = &nd_region->dev;
2071         dev->groups = nd_namespace_attribute_groups;
2072
2073         return &nsblk->common.dev;
2074 }
2075
2076 static struct device *nd_namespace_pmem_create(struct nd_region *nd_region)
2077 {
2078         struct nd_namespace_pmem *nspm;
2079         struct resource *res;
2080         struct device *dev;
2081
2082         if (!is_memory(&nd_region->dev))
2083                 return NULL;
2084
2085         nspm = kzalloc(sizeof(*nspm), GFP_KERNEL);
2086         if (!nspm)
2087                 return NULL;
2088
2089         dev = &nspm->nsio.common.dev;
2090         dev->type = &namespace_pmem_device_type;
2091         dev->parent = &nd_region->dev;
2092         res = &nspm->nsio.res;
2093         res->name = dev_name(&nd_region->dev);
2094         res->flags = IORESOURCE_MEM;
2095
2096         nspm->id = ida_simple_get(&nd_region->ns_ida, 0, 0, GFP_KERNEL);
2097         if (nspm->id < 0) {
2098                 kfree(nspm);
2099                 return NULL;
2100         }
2101         dev_set_name(dev, "namespace%d.%d", nd_region->id, nspm->id);
2102         dev->groups = nd_namespace_attribute_groups;
2103         nd_namespace_pmem_set_resource(nd_region, nspm, 0);
2104
2105         return dev;
2106 }
2107
2108 void nd_region_create_ns_seed(struct nd_region *nd_region)
2109 {
2110         WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
2111
2112         if (nd_region_to_nstype(nd_region) == ND_DEVICE_NAMESPACE_IO)
2113                 return;
2114
2115         if (is_nd_blk(&nd_region->dev))
2116                 nd_region->ns_seed = nd_namespace_blk_create(nd_region);
2117         else
2118                 nd_region->ns_seed = nd_namespace_pmem_create(nd_region);
2119
2120         /*
2121          * Seed creation failures are not fatal, provisioning is simply
2122          * disabled until memory becomes available
2123          */
2124         if (!nd_region->ns_seed)
2125                 dev_err(&nd_region->dev, "failed to create %s namespace\n",
2126                                 is_nd_blk(&nd_region->dev) ? "blk" : "pmem");
2127         else
2128                 nd_device_register(nd_region->ns_seed);
2129 }
2130
2131 void nd_region_create_dax_seed(struct nd_region *nd_region)
2132 {
2133         WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
2134         nd_region->dax_seed = nd_dax_create(nd_region);
2135         /*
2136          * Seed creation failures are not fatal, provisioning is simply
2137          * disabled until memory becomes available
2138          */
2139         if (!nd_region->dax_seed)
2140                 dev_err(&nd_region->dev, "failed to create dax namespace\n");
2141 }
2142
2143 void nd_region_create_pfn_seed(struct nd_region *nd_region)
2144 {
2145         WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
2146         nd_region->pfn_seed = nd_pfn_create(nd_region);
2147         /*
2148          * Seed creation failures are not fatal, provisioning is simply
2149          * disabled until memory becomes available
2150          */
2151         if (!nd_region->pfn_seed)
2152                 dev_err(&nd_region->dev, "failed to create pfn namespace\n");
2153 }
2154
2155 void nd_region_create_btt_seed(struct nd_region *nd_region)
2156 {
2157         WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
2158         nd_region->btt_seed = nd_btt_create(nd_region);
2159         /*
2160          * Seed creation failures are not fatal, provisioning is simply
2161          * disabled until memory becomes available
2162          */
2163         if (!nd_region->btt_seed)
2164                 dev_err(&nd_region->dev, "failed to create btt namespace\n");
2165 }
2166
2167 static int add_namespace_resource(struct nd_region *nd_region,
2168                 struct nd_namespace_label *nd_label, struct device **devs,
2169                 int count)
2170 {
2171         struct nd_mapping *nd_mapping = &nd_region->mapping[0];
2172         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
2173         int i;
2174
2175         for (i = 0; i < count; i++) {
2176                 u8 *uuid = namespace_to_uuid(devs[i]);
2177                 struct resource *res;
2178
2179                 if (IS_ERR_OR_NULL(uuid)) {
2180                         WARN_ON(1);
2181                         continue;
2182                 }
2183
2184                 if (memcmp(uuid, nd_label->uuid, NSLABEL_UUID_LEN) != 0)
2185                         continue;
2186                 if (is_namespace_blk(devs[i])) {
2187                         res = nsblk_add_resource(nd_region, ndd,
2188                                         to_nd_namespace_blk(devs[i]),
2189                                         __le64_to_cpu(nd_label->dpa));
2190                         if (!res)
2191                                 return -ENXIO;
2192                         nd_dbg_dpa(nd_region, ndd, res, "%d assign\n", count);
2193                 } else {
2194                         dev_err(&nd_region->dev,
2195                                         "error: conflicting extents for uuid: %pUb\n",
2196                                         nd_label->uuid);
2197                         return -ENXIO;
2198                 }
2199                 break;
2200         }
2201
2202         return i;
2203 }
2204
2205 static struct device *create_namespace_blk(struct nd_region *nd_region,
2206                 struct nd_namespace_label *nd_label, int count)
2207 {
2208
2209         struct nd_mapping *nd_mapping = &nd_region->mapping[0];
2210         struct nd_interleave_set *nd_set = nd_region->nd_set;
2211         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
2212         struct nd_namespace_blk *nsblk;
2213         char name[NSLABEL_NAME_LEN];
2214         struct device *dev = NULL;
2215         struct resource *res;
2216
2217         if (namespace_label_has(ndd, type_guid)) {
2218                 if (!guid_equal(&nd_set->type_guid, &nd_label->type_guid)) {
2219                         dev_dbg(ndd->dev, "expect type_guid %pUb got %pUb\n",
2220                                         nd_set->type_guid.b,
2221                                         nd_label->type_guid.b);
2222                         return ERR_PTR(-EAGAIN);
2223                 }
2224
2225                 if (nd_label->isetcookie != __cpu_to_le64(nd_set->cookie2)) {
2226                         dev_dbg(ndd->dev, "expect cookie %#llx got %#llx\n",
2227                                         nd_set->cookie2,
2228                                         __le64_to_cpu(nd_label->isetcookie));
2229                         return ERR_PTR(-EAGAIN);
2230                 }
2231         }
2232
2233         nsblk = kzalloc(sizeof(*nsblk), GFP_KERNEL);
2234         if (!nsblk)
2235                 return ERR_PTR(-ENOMEM);
2236         dev = &nsblk->common.dev;
2237         dev->type = &namespace_blk_device_type;
2238         dev->parent = &nd_region->dev;
2239         nsblk->id = -1;
2240         nsblk->lbasize = __le64_to_cpu(nd_label->lbasize);
2241         nsblk->uuid = kmemdup(nd_label->uuid, NSLABEL_UUID_LEN,
2242                         GFP_KERNEL);
2243         if (namespace_label_has(ndd, abstraction_guid))
2244                 nsblk->common.claim_class
2245                         = to_nvdimm_cclass(&nd_label->abstraction_guid);
2246         if (!nsblk->uuid)
2247                 goto blk_err;
2248         memcpy(name, nd_label->name, NSLABEL_NAME_LEN);
2249         if (name[0])
2250                 nsblk->alt_name = kmemdup(name, NSLABEL_NAME_LEN,
2251                                 GFP_KERNEL);
2252         res = nsblk_add_resource(nd_region, ndd, nsblk,
2253                         __le64_to_cpu(nd_label->dpa));
2254         if (!res)
2255                 goto blk_err;
2256         nd_dbg_dpa(nd_region, ndd, res, "%d: assign\n", count);
2257         return dev;
2258  blk_err:
2259         namespace_blk_release(dev);
2260         return ERR_PTR(-ENXIO);
2261 }
2262
2263 static int cmp_dpa(const void *a, const void *b)
2264 {
2265         const struct device *dev_a = *(const struct device **) a;
2266         const struct device *dev_b = *(const struct device **) b;
2267         struct nd_namespace_blk *nsblk_a, *nsblk_b;
2268         struct nd_namespace_pmem *nspm_a, *nspm_b;
2269
2270         if (is_namespace_io(dev_a))
2271                 return 0;
2272
2273         if (is_namespace_blk(dev_a)) {
2274                 nsblk_a = to_nd_namespace_blk(dev_a);
2275                 nsblk_b = to_nd_namespace_blk(dev_b);
2276
2277                 return memcmp(&nsblk_a->res[0]->start, &nsblk_b->res[0]->start,
2278                                 sizeof(resource_size_t));
2279         }
2280
2281         nspm_a = to_nd_namespace_pmem(dev_a);
2282         nspm_b = to_nd_namespace_pmem(dev_b);
2283
2284         return memcmp(&nspm_a->nsio.res.start, &nspm_b->nsio.res.start,
2285                         sizeof(resource_size_t));
2286 }
2287
2288 static struct device **scan_labels(struct nd_region *nd_region)
2289 {
2290         int i, count = 0;
2291         struct device *dev, **devs = NULL;
2292         struct nd_label_ent *label_ent, *e;
2293         struct nd_mapping *nd_mapping = &nd_region->mapping[0];
2294         resource_size_t map_end = nd_mapping->start + nd_mapping->size - 1;
2295
2296         /* "safe" because create_namespace_pmem() might list_move() label_ent */
2297         list_for_each_entry_safe(label_ent, e, &nd_mapping->labels, list) {
2298                 struct nd_namespace_label *nd_label = label_ent->label;
2299                 struct device **__devs;
2300                 u32 flags;
2301
2302                 if (!nd_label)
2303                         continue;
2304                 flags = __le32_to_cpu(nd_label->flags);
2305                 if (is_nd_blk(&nd_region->dev)
2306                                 == !!(flags & NSLABEL_FLAG_LOCAL))
2307                         /* pass, region matches label type */;
2308                 else
2309                         continue;
2310
2311                 /* skip labels that describe extents outside of the region */
2312                 if (nd_label->dpa < nd_mapping->start || nd_label->dpa > map_end)
2313                         continue;
2314
2315                 i = add_namespace_resource(nd_region, nd_label, devs, count);
2316                 if (i < 0)
2317                         goto err;
2318                 if (i < count)
2319                         continue;
2320                 __devs = kcalloc(count + 2, sizeof(dev), GFP_KERNEL);
2321                 if (!__devs)
2322                         goto err;
2323                 memcpy(__devs, devs, sizeof(dev) * count);
2324                 kfree(devs);
2325                 devs = __devs;
2326
2327                 if (is_nd_blk(&nd_region->dev))
2328                         dev = create_namespace_blk(nd_region, nd_label, count);
2329                 else {
2330                         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
2331                         struct nd_namespace_index *nsindex;
2332
2333                         nsindex = to_namespace_index(ndd, ndd->ns_current);
2334                         dev = create_namespace_pmem(nd_region, nsindex, nd_label);
2335                 }
2336
2337                 if (IS_ERR(dev)) {
2338                         switch (PTR_ERR(dev)) {
2339                         case -EAGAIN:
2340                                 /* skip invalid labels */
2341                                 continue;
2342                         case -ENODEV:
2343                                 /* fallthrough to seed creation */
2344                                 break;
2345                         default:
2346                                 goto err;
2347                         }
2348                 } else
2349                         devs[count++] = dev;
2350
2351         }
2352
2353         dev_dbg(&nd_region->dev, "discovered %d %s namespace%s\n",
2354                         count, is_nd_blk(&nd_region->dev)
2355                         ? "blk" : "pmem", count == 1 ? "" : "s");
2356
2357         if (count == 0) {
2358                 /* Publish a zero-sized namespace for userspace to configure. */
2359                 nd_mapping_free_labels(nd_mapping);
2360
2361                 devs = kcalloc(2, sizeof(dev), GFP_KERNEL);
2362                 if (!devs)
2363                         goto err;
2364                 if (is_nd_blk(&nd_region->dev)) {
2365                         struct nd_namespace_blk *nsblk;
2366
2367                         nsblk = kzalloc(sizeof(*nsblk), GFP_KERNEL);
2368                         if (!nsblk)
2369                                 goto err;
2370                         dev = &nsblk->common.dev;
2371                         dev->type = &namespace_blk_device_type;
2372                 } else {
2373                         struct nd_namespace_pmem *nspm;
2374
2375                         nspm = kzalloc(sizeof(*nspm), GFP_KERNEL);
2376                         if (!nspm)
2377                                 goto err;
2378                         dev = &nspm->nsio.common.dev;
2379                         dev->type = &namespace_pmem_device_type;
2380                         nd_namespace_pmem_set_resource(nd_region, nspm, 0);
2381                 }
2382                 dev->parent = &nd_region->dev;
2383                 devs[count++] = dev;
2384         } else if (is_memory(&nd_region->dev)) {
2385                 /* clean unselected labels */
2386                 for (i = 0; i < nd_region->ndr_mappings; i++) {
2387                         struct list_head *l, *e;
2388                         LIST_HEAD(list);
2389                         int j;
2390
2391                         nd_mapping = &nd_region->mapping[i];
2392                         if (list_empty(&nd_mapping->labels)) {
2393                                 WARN_ON(1);
2394                                 continue;
2395                         }
2396
2397                         j = count;
2398                         list_for_each_safe(l, e, &nd_mapping->labels) {
2399                                 if (!j--)
2400                                         break;
2401                                 list_move_tail(l, &list);
2402                         }
2403                         nd_mapping_free_labels(nd_mapping);
2404                         list_splice_init(&list, &nd_mapping->labels);
2405                 }
2406         }
2407
2408         if (count > 1)
2409                 sort(devs, count, sizeof(struct device *), cmp_dpa, NULL);
2410
2411         return devs;
2412
2413  err:
2414         if (devs) {
2415                 for (i = 0; devs[i]; i++)
2416                         if (is_nd_blk(&nd_region->dev))
2417                                 namespace_blk_release(devs[i]);
2418                         else
2419                                 namespace_pmem_release(devs[i]);
2420                 kfree(devs);
2421         }
2422         return NULL;
2423 }
2424
2425 static struct device **create_namespaces(struct nd_region *nd_region)
2426 {
2427         struct nd_mapping *nd_mapping;
2428         struct device **devs;
2429         int i;
2430
2431         if (nd_region->ndr_mappings == 0)
2432                 return NULL;
2433
2434         /* lock down all mappings while we scan labels */
2435         for (i = 0; i < nd_region->ndr_mappings; i++) {
2436                 nd_mapping = &nd_region->mapping[i];
2437                 mutex_lock_nested(&nd_mapping->lock, i);
2438         }
2439
2440         devs = scan_labels(nd_region);
2441
2442         for (i = 0; i < nd_region->ndr_mappings; i++) {
2443                 int reverse = nd_region->ndr_mappings - 1 - i;
2444
2445                 nd_mapping = &nd_region->mapping[reverse];
2446                 mutex_unlock(&nd_mapping->lock);
2447         }
2448
2449         return devs;
2450 }
2451
2452 static int init_active_labels(struct nd_region *nd_region)
2453 {
2454         int i;
2455
2456         for (i = 0; i < nd_region->ndr_mappings; i++) {
2457                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
2458                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
2459                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
2460                 struct nd_label_ent *label_ent;
2461                 int count, j;
2462
2463                 /*
2464                  * If the dimm is disabled then we may need to prevent
2465                  * the region from being activated.
2466                  */
2467                 if (!ndd) {
2468                         if (test_bit(NDD_LOCKED, &nvdimm->flags))
2469                                 /* fail, label data may be unreadable */;
2470                         else if (test_bit(NDD_ALIASING, &nvdimm->flags))
2471                                 /* fail, labels needed to disambiguate dpa */;
2472                         else
2473                                 return 0;
2474
2475                         dev_err(&nd_region->dev, "%s: is %s, failing probe\n",
2476                                         dev_name(&nd_mapping->nvdimm->dev),
2477                                         test_bit(NDD_LOCKED, &nvdimm->flags)
2478                                         ? "locked" : "disabled");
2479                         return -ENXIO;
2480                 }
2481                 nd_mapping->ndd = ndd;
2482                 atomic_inc(&nvdimm->busy);
2483                 get_ndd(ndd);
2484
2485                 count = nd_label_active_count(ndd);
2486                 dev_dbg(ndd->dev, "count: %d\n", count);
2487                 if (!count)
2488                         continue;
2489                 for (j = 0; j < count; j++) {
2490                         struct nd_namespace_label *label;
2491
2492                         label_ent = kzalloc(sizeof(*label_ent), GFP_KERNEL);
2493                         if (!label_ent)
2494                                 break;
2495                         label = nd_label_active(ndd, j);
2496                         label_ent->label = label;
2497
2498                         mutex_lock(&nd_mapping->lock);
2499                         list_add_tail(&label_ent->list, &nd_mapping->labels);
2500                         mutex_unlock(&nd_mapping->lock);
2501                 }
2502
2503                 if (j >= count)
2504                         continue;
2505
2506                 mutex_lock(&nd_mapping->lock);
2507                 nd_mapping_free_labels(nd_mapping);
2508                 mutex_unlock(&nd_mapping->lock);
2509                 return -ENOMEM;
2510         }
2511
2512         return 0;
2513 }
2514
2515 int nd_region_register_namespaces(struct nd_region *nd_region, int *err)
2516 {
2517         struct device **devs = NULL;
2518         int i, rc = 0, type;
2519
2520         *err = 0;
2521         nvdimm_bus_lock(&nd_region->dev);
2522         rc = init_active_labels(nd_region);
2523         if (rc) {
2524                 nvdimm_bus_unlock(&nd_region->dev);
2525                 return rc;
2526         }
2527
2528         type = nd_region_to_nstype(nd_region);
2529         switch (type) {
2530         case ND_DEVICE_NAMESPACE_IO:
2531                 devs = create_namespace_io(nd_region);
2532                 break;
2533         case ND_DEVICE_NAMESPACE_PMEM:
2534         case ND_DEVICE_NAMESPACE_BLK:
2535                 devs = create_namespaces(nd_region);
2536                 break;
2537         default:
2538                 break;
2539         }
2540         nvdimm_bus_unlock(&nd_region->dev);
2541
2542         if (!devs)
2543                 return -ENODEV;
2544
2545         for (i = 0; devs[i]; i++) {
2546                 struct device *dev = devs[i];
2547                 int id;
2548
2549                 if (type == ND_DEVICE_NAMESPACE_BLK) {
2550                         struct nd_namespace_blk *nsblk;
2551
2552                         nsblk = to_nd_namespace_blk(dev);
2553                         id = ida_simple_get(&nd_region->ns_ida, 0, 0,
2554                                         GFP_KERNEL);
2555                         nsblk->id = id;
2556                 } else if (type == ND_DEVICE_NAMESPACE_PMEM) {
2557                         struct nd_namespace_pmem *nspm;
2558
2559                         nspm = to_nd_namespace_pmem(dev);
2560                         id = ida_simple_get(&nd_region->ns_ida, 0, 0,
2561                                         GFP_KERNEL);
2562                         nspm->id = id;
2563                 } else
2564                         id = i;
2565
2566                 if (id < 0)
2567                         break;
2568                 dev_set_name(dev, "namespace%d.%d", nd_region->id, id);
2569                 dev->groups = nd_namespace_attribute_groups;
2570                 nd_device_register(dev);
2571         }
2572         if (i)
2573                 nd_region->ns_seed = devs[0];
2574
2575         if (devs[i]) {
2576                 int j;
2577
2578                 for (j = i; devs[j]; j++) {
2579                         struct device *dev = devs[j];
2580
2581                         device_initialize(dev);
2582                         put_device(dev);
2583                 }
2584                 *err = j - i;
2585                 /*
2586                  * All of the namespaces we tried to register failed, so
2587                  * fail region activation.
2588                  */
2589                 if (*err == 0)
2590                         rc = -ENODEV;
2591         }
2592         kfree(devs);
2593
2594         if (rc == -ENODEV)
2595                 return rc;
2596
2597         return i;
2598 }